Fireworks AI serverless token pricing + serverless GPU hourly rates (2026-10-04 list price)

TWO UNITS IN ONE RECORD, DO NOT MIX THEM. (a) USD per 1M TOKENS, serverless, list price: models up to 16B parameters $0.50 input / $1.00 cached / $1.00 output / $2.00 output-cached; 16.1B-80B $3.00 / $6.00 / $6.00 / $12.00; 80B-300B (e.g. Qwen3-235B, gpt-oss-120B) $6.00 / $12.00 / $12.00 / $24.00; over 300B (e.g. DeepSeek V3, Kimi K2) $10.00 / $20.00 / $20.00 / $40.00. (b) USD per GPU-HOUR for the same company's serverless GPU product: H100 80GB $0.134/min = $8.00/hr and H200 141GB $0.134/min = $8.00/hr; B200 180GB $0.217/min = $13.00/hr; B300 288GB $0.250/min = $15.00/hr; GB300 288GB $0.334/min = $20.00/hr. price_usd = 80B-300B tier serverless input. OBSERVATION DATE: the figure above was observed/listed on 2026-10-04 unless a different date is stated inline; treat it as needing re-verification once it is more than a quarter old PERISHABLE-FIGURE STAMP (added 2026-10-04, uniform across data/supply/ so it is greppable). PRICE CLASS: PUBLISHED LIST PRICE - an asking rate a new customer is quoted, NOT a realised, discounted or negotiated transaction rate. A real bill is at or below this and typically well below it once commitments, credits and committed-use discounts apply. OBSERVED: 2026-10-04 (the date this record was verified against its source; the per-figure fetch date is not restated in price_basis) RE-VERIFY BY: 2027-01-04 - the repo's quarterly re-verification bar. After that date this figure is presumed drifted and must be re-checked against its source before it is quoted again. A price without its observation date is noise; this stamp exists so a reader can tell a fresh figure from a stale one without reading prose. UNIT: TWO UNITS IN ONE RECORD, DO NOT MIX THEM. (a) USD per 1M TOKENS, serverless, list price: models up to 16B parameters $0.50 input / $1.00 cached / $1.00 output / $2.00 output READING RULE: MAJOR UNIT WARNING: this record carries TWO UNIT SYSTEMS at once (per-1M-tokens and per-GPU-hour). price_usd is the token figure; the GPU-hour figures belong to a different product and must never be mixed with it.

lead_time_weeks is null because NO NUMBER APPLIES, and the reason is specific rather than generic: this record prices a metered service term (tokens, egress, storage) rather than hardware capacity, so there is no order-to-delivery interval to report at all. Calling this 'no published lead time' would be wrong in the other direction - nothing was withheld.

no EAR citation: no controlled item is transferred by this transaction; not an EAR citation: provider service terms and end-use policy (the AI Diffusion Rule briefly tiered service access before its 2025-05-13 rescission)

MECHANISM: a token-priced API, so no accelerator crosses a border and the 3A090 hardware licence regime does not attach to the transaction. Export control becomes an END-USER AND TERMS question instead: the provider applies geographic and end-use restrictions to service access, and a customer can still be refused under sanctions or end-use policy wherever the compute sits. A reader targeting a restricted destination should read the provider's terms, not the EAR.

This record deliberately carries both units because Fireworks is the one provider in this set that sells the SAME hardware two ways on one page, and the relationship between the two prices is the most useful thing a reader can learn from it. $8.00 per H100-hour divided by a 1M-token rate of, say, $12.00 for an 80-300B model implies break-even at roughly 666,667 output tokens per GPU-hour - and against the >300B rate of $40.00 it is 5,000,000 tokens per GPU-hour. Compare that with a real decode ceiling of a few thousand tok/s: a single H100 at 2,000 tok/s for a full hour produces about 7.2M output tokens, so the per-token serverless rate is genuinely competitive only for dense high-throughput serving, and the per-GPU rate wins as soon as you can keep the card busy. That crossover is the whole serverless-vs-GPU decision and it moves with your achieved tok/s, not with the list price. Note also Fireworks prices per-PARAMETER-SIZE BAND rather than per named model, so a new release inside a band is priced by the band automatically. EXPORT CONTROL (2026-10-04): NOT APPLICABLE IN THE HARDWARE-SHIPMENT SENSE - this record prices a token-based API, so no accelerator is transferred across a border and the ECCN 3A090 hardware licence regime does not attach to the transaction. Recorded null rather than false precisely because a token API is where export control instead becomes an END-USER/TERMS question: the provider applies geographic and end-user restrictions to service access (the AI Diffusion Rule briefly put a tiered country structure on exactly this kind of service before being rescinded on 2025-05-13), and a customer can still be refused under sanctions or end-use policy regardless of where the compute sits. Anyone reading this record for a restricted destination should look at the provider's terms, not at the EAR. Sources for the regime: spl-bis-2025-05-ai-diffusion-rescission, spl-bis-2026-05-d5-guidance. EXPORT CONTROL (2026-10-04): NOT APPLICABLE IN THE HARDWARE-SHIPMENT SENSE - this record prices a token-based API, so no accelerator crosses a border and the ECCN 3A090 hardware licence regime does not attach to the transaction. Recorded NULL rather than false precisely because a token API is where export control instead becomes an END-USER and TERMS question: the provider applies geographic and end-user restrictions to service access (the AI Diffusion Rule briefly put a tiered country structure on exactly this kind of service before being rescinded and unenforced from 2025-05-13), and a customer can still be refused under sanctions or end-use policy regardless of where the compute sits. For a restricted destination, read the provider's terms, not the EAR. Sources: spl-bis-2025-05-ai-diffusion-rescission, spl-bis-2026-05-d5-guidance. LEAD TIME: null - correct and not a gap. A token-priced serverless API is provisioned on request, so there is no lead time to publish; the binding constraint when a model is at capacity is provider-side queuing and any per-account concurrency limit, which this repo records in notes where a vendor publishes one (see the DeepSeek concurrency figures already here). Deliberately NOT recorded: any 'model capacity lead time' estimate, because vendors publish none and the SEO 'GPU lead time' numbers that circulate do not apply to a rented endpoint. Probe log in source spl-negative-no-oem-publishes-lead-time. ACCELERATOR LINK ADDED 2026-10-04: Fireworks is the one provider in this slice selling the SAME controlled hardware two ways on one page - serverless per-token and a serverless per-GPU-hour product covering H100 80GB, H200 141GB, B200 180GB, B300 288GB and GB300 - so the ids link the two price records to the same silicon and make the break-even arithmetic in notes checkable.